An organization publishes an AI policy, creates a review board, adds a list of prohibited uses, and announces that employees may begin experimenting responsibly.
Six months later, several different outcomes may appear:
- Some employees avoid approved tools because they cannot tell what is permitted.
- Others use public tools quietly because the approved path feels too slow.
- Managers interpret every AI use as high risk and escalate routine questions.
- Product teams involve governance only after a design is nearly complete.
- Leaders report adoption by counting licences or logins rather than appropriate use.
The policy exists. Governance exists. Adoption may still be unsafe, weak, or invisible.
This is not necessarily a conflict between governance and innovation. It is often an information-to-action problem:
People must recognize when a rule applies, understand why it matters, distinguish among risk levels, and complete the appropriate next step without inventing their own process.
AI governance is a system, not a warning message
The NIST AI Risk Management Framework describes governance as a cross-cutting function that should operate throughout the AI lifecycle. That matters for communication design. A policy page cannot carry the whole burden.
Employees encounter governance through:
- Procurement and access requests
- Product requirements and design reviews
- Data-handling controls
- Model and vendor documentation
- Training and examples
- Approval workflows
- Interface warnings
- Incident and escalation routes
- Manager expectations
Each touchpoint can make the intended action easier to recognize—or add another unsupported jump.
A broad statement such as “Do not put confidential information into unapproved AI tools” may be accurate. It does not tell a product manager whether a particular document is confidential, which approved tool fits the task, or what to do when none does.
The problem is not solved by making the warning louder. The path needs usable distinctions.
Why safety language can stall appropriate adoption
Governance communication often begins from the expert sender’s model:
- Identify every material risk.
- Express the controls precisely.
- Direct people to the authoritative policy.
- Require escalation when uncertain.
That sequence protects accuracy, but it may impose high Strain on the receiver.
The employee’s active questions are usually more local:
- Can I use this tool for this task?
- Which information can I enter?
- Must a person review the output?
- Can this result be sent to a customer?
- What record must I retain?
- Who can answer before my deadline?
When a policy answers at the wrong level, people must translate organizational language into a live decision. Some will stop. Some will guess. Some will route every question to a small governance team.
Low adoption is therefore ambiguous. It may indicate:
- A reasonable decision not to use AI
- Lack of a valuable use case
- Inability to access the approved system
- Uncertainty about permitted use
- Fear of surveillance or performance consequences
- Low confidence in output quality
- A review process that exceeds the value of the task
- Weak manager support
Treating all of these as “resistance” prevents a useful diagnosis.
Translate policy into role-specific decisions
A usable governance path starts with the receiver’s decision—not the policy table of contents.
Consider three receivers:
An employee using an approved assistant
They need task examples, data boundaries, verification expectations, and a quick route for uncertainty.
A product manager adding an AI feature
They need to know when governance review begins, which evidence is required, who owns the decision, and what changes at each risk level.
A governance reviewer
They need a bounded claim, intended users, data flow, failure modes, evaluation evidence, accountable owner, and deployment conditions.
The governing principles may be shared, but the entry points and required evidence differ.
This is receiver-aware governance. It does not weaken the rule. It makes the rule usable at the point of decision.
Design the path before, during, and after use
Human–AI interaction research has long emphasized that people need help forming accurate expectations, understanding system capabilities, correcting errors, and learning from behaviour over time. The Guidelines for Human-AI Interaction synthesize and test 18 such guidelines across AI-infused products.
For governance communication, the sequence can be organized into three moments.
Before use
Help the receiver determine:
- Whether the task is an approved use
- Which data and tools are allowed
- What the system can and cannot be trusted to do
- Which human remains accountable
- Whether a review is required before use begins
During use
Make the relevant control available in the workflow:
- Warn at the point sensitive data could be entered
- Show source and uncertainty where a judgment is made
- Require review when the consequence warrants it
- Make correction and escalation easy
- Preserve a record appropriate to the risk
After use
Support learning and accountability:
- Capture incidents and near misses
- Distinguish tool failure from process failure
- Review unequal or harmful effects
- Update examples when the system or policy changes
- Tell users what changed and why
The policy becomes a connected path rather than a document employees must repeatedly reinterpret.
Separate adoption from compliance and value
“Adoption” is often measured with one number: licences activated, weekly users, prompts submitted, or features enabled.
Those measures do not establish appropriate use.
A stronger measurement model separates:
| Transition | Example measure |
|---|---|
| Awareness | Can the person locate the applicable rule? |
| Classification | Can they distinguish approved, restricted, and prohibited use? |
| Capability | Can they complete an approved task with required verification? |
| Appropriate use | Is the tool used when it fits—and avoided when it does not? |
| Governance performance | Are reviews timely, consistent, and evidence-based? |
| Downstream value | Does the use improve quality, time, access, or another intended outcome? |
| Guardrails | What errors, incidents, inequities, workarounds, or burdens appear? |
A governance program should not maximize use. It should increase the probability of appropriate, accountable use while making reasonable refusal possible.
Audit governance communication with the Violet 5S™ method
The five cues provide a practical check.
Spotlite
Does the receiver see the decision that applies now, or only a long list of institutional priorities?
Substance
Are there concrete examples, definitions, evidence requirements, and boundaries—or only principles?
Source
Is ownership visible? Can the receiver tell who made the rule, who interprets it, and who remains accountable?
Strain
How many pages, forms, approvals, unfamiliar terms, and handoffs separate the question from a responsible answer?
Stake
Does the communication explain the receiver’s legitimate reason to follow the path: safer work, faster approval, clearer accountability, protected customers, or confidence to use an approved tool?
The answer is not automatically less governance. It may be clearer classification, earlier review, better examples, or a more direct decision route.
A practical governance communication checklist
Before announcing or revising an AI governance program, ask:
- Which receivers must make which recurring decisions?
- What should each person notice first?
- Which terms require examples?
- What is approved, restricted, and prohibited?
- Which human remains accountable at each step?
- Where does the control appear in the real workflow?
- How quickly can a bounded question receive an answer?
- What happens when the policy, tool, or risk changes?
- Which measures distinguish awareness, appropriate use, and value?
- Which guardrails detect harm, workarounds, and excessive burden?
If the organization cannot answer those questions, the next intervention may not be another training campaign. It may be redesigning the governance path.
Make governance confidence infrastructure
Good governance can increase confidence because people know what is allowed, what evidence is required, and where accountability sits.
Poorly translated governance can produce either paralysis or invisible workarounds.
The difference is not tone alone. It is whether the information system supports the decisions people must make.
For a related product perspective, read Building Receiver-Aware AI. To see the framework applied to organizational change, explore the AI workplace announcement case.
If an AI policy, approval flow, training sequence, or rollout message is not producing appropriate use, request a consultation for the wider system or send one message for a free attention audit.